Structure Logic: A New Method that Enables Richer Meaning Representations
Publication Date: 2014-Mar-21
The IP.com Prior Art Database
Within the area of natural language understanding there are a number of methods, formal and informal, for meaning representation. These approaches include first order logic, lambda calculus, and frame-based approaches. Meaning representation is important for semantic parsing and for other tasks of natural language understanding such as summarization. This paper introduces a new method for meaning representation called structure logic. Structure logic addresses an important weakness that results from using other formal meaning representation approaches: existing approaches are expressive but the very nature of their expressive capabilities often results in meaning representations that in themselves are not as structured as could be attained. Structure logic addresses this challenge via a set of underlying ontological commitments and by means of its representational constructs. The development of structure logic is a work in progress; however this approach has been successfully used in an experimental natural language understanding system that comprehends news stories. This disclosure also includes information about the ModelBuilder natural language understanding system. Finally, this disclosure includes C++ source code for the processor for the Star language, and it includes a set of Infopedia definitions.
Structure logic is introduced as a meaning representation approach that is more effective than existing approaches for fulfilling the objectives for structured information. A meaning representation is defined as a structured, cohesive and comprehensive representational artifact that is the output of a process of natural language understanding or comprehension as this process is applied to a sentence, fragment, or document of human natural language text. Existing meaning representation approaches include first order logics, semantic networks, and frame-based approaches (Jurafski and Martin, 2009). Sowa (2000) describes an approach known as conceptual graphs. From the semantic parsing side, lambda calculus is also used. Description logic is another alternative.
The main objective of such natural language understanding meaning representation techniques and languages is that of representing the original information in a structured way. Some such techniques are formal while others are informal. Structured representations support a variety of objectives: these objectives include indexed storage, which enables query and analysis, automated reasoning, for instance deductive inference, and other uses such as marking corpora. The objectives of question answering systems involve the transformation of questions into a structured form. A requirement for all structured representation approaches is that of comprehensiveness or completeness – the meaning representation language or logic must be capable of consistent and full coverage of the information in the input text. Each of the well-known approaches has strengths and weakness. Structure logic provides the infrastructure and a set of implicit guidelines that help to produce meaning representation artifacts that are highly structured in ways that are difficult to attain with existing approaches.
This paper describes a set of goals, and the features of structure logic that accomplish these objectives. Where comparisons are needed structure logic is contrasted with first order logic, since first order logic has a long-established position as a standard in the field of logic.
First order logic (FOL) is capable of comprehensive representation, but the unrestricted expressive capabilities of FOL often result in meaning representations that are in actuality unstructured. In contrast, structure logic facilitates and promotes practices that result in highly structured representations. Structure logic unabashedly makes use of explicit assumptions about the nature of the reality that is represented (the represented world) in order to achieve a higher level of organization than is typically achieved using a method such as FOL.
1.2 Structure is represented
Structure logic facilitates the creation of meaning representations that represent dimensional structure (usually but not always defined by the use of spatial and temporal dimensions). Propositional information is tied into a frame o...